ADir(p)NB: A cost-effective way to implement full map directory-based cache coherence protocols

Directories have been used to maintain cache coherency in shared memory mul tiprocessors with private caches. The traditional full map directory tracks the exact caching status for each shared memory block and is designed to b e efficient and simple, Unfortunately, the inherent directory size explosio n makes it unsuitable for large-scale multiprocessors. In this paper, we pr opose a new directory scheme, dubbed associative full map directory (ADir(p )NB) which reduces the directory storage requirement. The proposed ADir(p)N B uses one directory entry to maintain the sharing information for a set of exclusively cached memory blocks in a centralized linked list style. By im plementing dynamic cache pointer allocation, reclamation, and replacement h ints, ADir(p)NB can be implemented as "a full map directory with lower dire ctory memory cost." Our analysis indicates that, on a typical architectural paradigm, ADir(p)NB reduces memory overhead of a traditional full map dire ctory by up to 70-80 percent. In addition to the low memory overhead, we sh ow that the proposed scheme can be implemented with appropriate protocol mo dification and hardware addition. Simulation studies indicate that ADir(p)N B can achieve a competitive performance with the Dir(p)NB. Compared with li mited directory schemes, ADir(p)NB shows more stable and robust performance results on applications across a spectrum of memory sharing and access pat terns due to the elimination of directory overflows. We believe that ADir(p )NB can be employed as a design alternative of full map directory for moder ately large-scale and fine-grain shared memory multiprocessors.